Lab 1
Device Characterization
Measure the threshold, the square-law parameters, the subthreshold slope, the small-signal parameters, fT, the ZTC point and the corner spread of the core transistors.
Contents
| Lab 1 at a glance | |
|---|---|
| Goal | Measure the threshold, the square-law parameters, the subthreshold slope, the small-signal parameters, fT, the ZTC point and the corner spread of the core transistors. |
| Reproduces | Chapter 4 (Sections 4.2–4.7, Worked example 4.1) and Chapter 5 (Sections 5.3–5.7, Figures 5.5–5.8) |
| Time | 3 hours |
| Files | labs/lab1: vth_tb, idvg_tb, smallsig_tb, temp_tb, corner_tb (.sch) per PDK; sqlaw_fit.py |
Hand calculation
Worked example 4.1 biased an NMOS at 50 µA. Do the same for the PMOS of W/L = 5/0.5 with the square-law parameters of Table 4.2 (SG13G2: µCox = 70 µA/V², \(V_{TH}\) = 0.34 V; SKY130: 60 µA/V², 0.94 V): find \(V_{OV} = \sqrt{2I_D/(\mu C_{ox}W/L)}\) and \(\lvert V_{GS}\rvert = \lvert V_{TH}\rvert + V_{OV}\). Then estimate \(g_m = 2I_D/V_{OV}\) for the NMOS reference device of Worked example 5.1 (VOV = 0.187 V and 0.216 V).
Steps
- Threshold. In
labs/lab1/sg13g2(orsky130, aftersak-pdk sky130A), run../../sim.sh vth_tb.sch. Four transistors sit side by side, three NMOS devices of W = 5 µm at Lmin, 0.5 and 2 µm and a 5/0.5 PMOS, each with its drain 50 mV from its source. TheNGSPICEblock finds the maximum of \(g_m\) and extends the tangent there to zero current, exactly the construction of Figure 4.2. The PMOS is turned upside down (source and bulk at ground, gate at \(-V_G\) through the controlled sourceEgp) so that one sweep drives all four. - Saturation, the square law and Worked example 4.1. Run
../../sim.sh idvg_tb.sch, thenpython3 ../sqlaw_fit.py. The testbench prints the gate voltage for 50 µA at L = 0.5 µm and at Lmin, the shift caused by 0.3 V of body bias (M3 has its bulk at −0.3 V) and the subthreshold slopes; the script fits \(\sqrt{I_D}\) against \(V_{GS}\) between 0.1 and 0.4 V above the threshold of step 1, prints µCox and the fitted \(V_{TH}\), and repeats your hand calculation with them. - Small-signal parameters three ways. Run
../../sim.sh smallsig_tb.sch. It finds the 50 µA bias, then extracts \(g_m\), \(g_{ds}\) and \(C_{gg}\) from the operating point, from numerical derivatives and from AC analyses at 1 MHz (Section 5.5), and finally measures \(f_T\) where the current gain \(h_{21} = i_d/i_g\) falls to one. In the GUI,plot db(h21)after the run shows Figure 5.5(a). - Temperature. Run
../../sim.sh temp_tb.sch. It sweeps the gate at −40, 27 and 125 °C and prints the gate voltage where the hot and cold curves cross (the ZTC point) and the change of the 50 µA current at fixed \(V_{GS}\). - Corners.
corner_tb.schholds both devices at the \(\lvert V_{GS}\rvert\) that gives 50 µA in tt (taken from step 2). Run it in all five corners with../../corners.py corner_tb.sch; divide each current by its tt value to get Figure 5.8.
Checkpoint
Table 18.2. Lab 1 results (W = 5 µm, tt, 27 °C unless stated). Saturation values at |VDS| = VDD/2.
| Quantity | SG13G2 | SKY130 |
|---|---|---|
| \(V_{TH}\) by extrapolation, NMOS Lmin / 0.5 / 2 µm | 0.516 / 0.311 / 0.250 V | 0.724 / 0.638 / 0.576 V |
| \(\lvert V_{TH}\rvert\) by extrapolation, PMOS 5/0.5 | 0.382 V | 0.979 V |
| Square-law fit µCox, \(V_{TH}\): NMOS / PMOS 5/0.5 | 286 µA/V², 0.249 V / 70 µA/V², 0.340 V | 215 µA/V², 0.595 V / 60 µA/V², 0.941 V |
| \(V_{GS}\) for 50 µA, NMOS 5/0.5: square law / simulated | 0.436 / 0.434 V | 0.810 / 0.811 V |
| \(\lvert V_{GS}\rvert\) for 50 µA, PMOS 5/0.5: square law / simulated | 0.718 / 0.718 V | 1.347 / 1.347 V |
| Body effect, \(\Delta V_{GS}\) for \(V_{SB}\) = 0.3 V; \(V_{GS}\) at Lmin | 35 mV; 0.516 V | 59 mV; 0.754 V |
| Subthreshold slope, NMOS / PMOS 5/0.5 | 74 / 74 mV/decade | 86 / 79 mV/decade |
| \(g_m\): op / DC derivative / AC | 499.5 / 498.7 / 499.5 µS | 473.2 / 473.2 / 473.2 µS |
| \(g_{ds}\): op / DC derivative / AC; \(C_{gg}\): op / AC | 18.91 / 18.95 / 18.91 µS; 20.97 / 20.97 fF | 4.529 / 4.529 / 4.529 µS; 16.02 / 16.02 fF |
| \(f_T\) from \(\lvert h_{21}\rvert\) = 1 / from \(g_m/(2\pi C_{gg})\) | 4.12 / 3.79 GHz | 5.02 / 4.70 GHz |
| ZTC point: \(V_{GS}\), \(I_D/W\) | 0.539 V, 23.5 µA/µm | 0.934 V, 24.7 µA/µm |
| 50 µA bias at fixed \(V_{GS}\): change at 125 °C / −40 °C | +23 % / −21 % | +20 % / −22 % |
| NMOS current in ss / ff, relative to tt (\(\Delta V_{TH}\)) | 0.57× / 1.57× (+47 / −46 mV) | 0.74× / 1.32× (+21 / −22 mV) |
sf corner: NMOS / PMOS current relative to tt |
0.77× / 1.15× | 1.44× / 0.70× |
The DC-derivative \(g_m\) of SG13G2 is 0.16 % low because smallsig_tb differentiates over only three points 1 mV apart; widen the sweep to 21 points, as Section 5.5 does, and it agrees with the other two to four digits. The leakage at \(V_{GS} = 0\) that idvg_tb also prints (ioff_n) is not in the table: at a fraction of a picoampere it depends on the simulator’s convergence settings as much as on the model.
Questions
- The SG13G2 NMOS threshold rises by 266 mV from L = 2 µm to Lmin, the SKY130 one by 148 mV. Which effect of Section 4.6 is this, and why does it make it dangerous to mirror a minimum-length device into a long one?
- Your PMOS hand value of step 2 agrees with the simulation to the millivolt, while the minimum-length NMOS of Worked example 4.1 was 172 mV off in SG13G2. What is different about the two predictions?
- \(g_m/(2\pi C_{gg})\) is 8 % (SG13G2) and 6 % (SKY130) below the \(f_T\) of the AC measurement. Which current, ignored by that definition, flows into the drain at high frequency?
- In SG13G2 the letters
sfmake the NMOS slow; in SKY130 they make it fast. How would you find out which is which for a third PDK, with one testbench and one run ofcorners.py?
Extension
Add a 5/0.5 PMOS to temp_tb.sch (upside down, as in idvg_tb) and find its ZTC point. Then change vth_tb.sch to the lengths 0.13, 0.25, 0.5, 1, 2 and 5 µm (0.15 µm instead of 0.13 µm in SKY130) and plot \(V_{TH}\) against L to reproduce Figure 4.7(a), including the PMOS that behaves differently in the two processes.